Rethinking Data for Intelligent Computing
summary
Argues intelligence equals free energy minimization and breaks perception, action, learning into hierarchical belief nodes that predict and explain asynchronous sensory streams. Describes a small Scala toy implementation of Hebbian learning inspired by NewPIC and Jeff Hawkins, aiming for rodent-level agents that learn online from unlabeled, noisy data rather than batch datasets. Emphasizes data-as-present, parallel sampling, use of existing CPUs for storage, and warns about non-determinism and lack of transparency.
AI-written summary · about
chapters (10)
- 0:00Introduction and context
- 2:58The free energy principle
- 5:54Mechanics of intelligence
- 11:55Hierarchical perception
- 23:51Action and goals
- 28:21Learning through experience
- 30:58Future of intelligent machines
- 35:23Capabilities and potential
- 39:58Redefining data for AI
- 44:24Challenges and ethics
related talks
46mDeveloping Intuition For AI As A Leader: Path To Data Literacy
41mModels Schm-odels: Why you Should Care about Data-Centric AI
11mBig Data Moonshots and Ground Control
37mLife of a data scientist
01h 00mData, visualization, and designing AI
30mData-Centric AI Cookbook-let's prep that data
23mData - UI rEvolution
19mData science demystified